12/3/2019

Initial Inquiries

  • Can we predict the time it takes for the Portland Police Bureau to respond to calls?
  • What can we say about the influence that race and wealth have on the reaction time of police?

Our Data

We combined two data sets: 2010 Census Bureau Data and 2012 Portland Police Bureau Data on Dispatched Calls and Portland Monthly income data. Our variables:
-Neighborhood
-Priority
-Percent of Population that's White
-Population Density
-Mean Income Our dependent variable: -Time in Queue

Portland

Choropleth Map of Peoples Below Poverty Level

Portland

Choropleth Map of Household Income by Range

Portland

Choropleth Map of Average Response Time

Portland

Choropleth Map of Number of Dispatches

Methods

Analysing Data

Approaches in Analysing Data:
- Bivariate Analysis
- PCA

Queue Time Frequency

Bivariate Analysis

Race and Priority

Bivariate Analysis

Population Density and Priority

Bivariate Analysis

Mean Income

PCA

Screeplot

Data Analysis

Initial Assessment

-Priority of Call

-Low Racial Diversity

Methods

Prediction Models

Attempted Models:

-Linear Regression

-Ridge Regression

-Lasso

-Regression Tree

-Boosted Tree

-Random Forest

Model Building

Linear Regression

Model Building

Linear Regression

Model Building

Linear Regression

Model Building

Linear Regression

Model Building with caretEnsemble

Model Building with caretEnsemble

Results

Results

Results

Results

Challenges

Dataset

  • size

  • skew

  • variations among models

  • not standardized between cities

Conclusion

Best Model: GBM Boosted Random Forest
Strongest Predictor: Income (excluding call priority and type of call)

Citation